1 · The thirty-one names, in the only order they are ever read
Pick one. Every word underneath it with a dotted line under it has an exact number printed against it by the Met Office: rough is 2.5 to 4 metres, gale 8 is 34 to 40 knots, soon means between six and twelve hours from the time of issue. The forecast is a natural language that was made into a formal one, and broadcast nightly to a country that mostly does not know it is hearing a specification.
Not every word has a number. Occasionally, at times, for a time and at first are load-bearing in almost every bulletin and are defined on no Met Office page and in no factsheet we could find; the only Met Office definitions we could find for them are in a conference presentation, which is a different kind of source and is not treated as the same thing. Perhaps and in places have no definition anywhere we looked. The dotted lines mark the words that do; their absence marks the words that do not.
2 · The order is a tour, and the tour is nearly the best one
Consecutive areas whose forecasts come out the same can be read together, which is why the broadcast says Viking, North Utsire, South Utsire and not the same sentence three times. So the order is doing a job: it decides how often that saving is available. Put the thirty-one areas in a random order and the same weather needs about six more lines to read out.
That makes the order an optimisation problem with an exact answer. The number of lines a bulletin needs is the number of runs of identical text along the order, and the expected number of runs is 1 + Σ P(neighbours differ) over the thirty consecutive pairs, additive exactly, by linearity of expectation. Minimising it is a maximum-weight Hamiltonian path on thirty-one nodes, which two independent integer-programming solvers settle to proven optimality in under a second.
Each area sits at the mean of the corner co-ordinates the Met Office prints for it in Factsheet 8, Table 1. These are not the areas: the published table gives seaward corners only and never describes the coastline, which also bounds most of them, so this is a chart of where the thirty-one names are, not a map of what they cover. The line is the order the names are read in.
| order | expected lines per bulletin | |
|---|---|---|
| a random order of the same thirty-one | ||
| the order they have read since 1949 | ||
| the best order that exists | proved, twice |
Swap two names and see
Click any two areas in the list below to exchange them. The number is the expected number of lines the bulletin would need, over every midnight bulletin in this archive, if the forecast were read in the order you have made.
3 · One bulletin, collapsed
Here is a single issue with its identical neighbours merged. The bar under it is the whole archive: one column per bulletin, one row per sea area, the colour the strongest Beaufort force that area's wind sentence commits to. Nineteen months of weather crossing the same thirty-one boxes in the same order. Drag it.
4 · What the broadcast actually does, which is more than this
Merging identical neighbours is the compression that loses nothing, and it is the only one an archive of per-area text can measure. The forecaster does something stronger. Areas are grouped when conditions are merely similar, and the difference is carried as a named qualifier inside the group (occasionally 7 in Thames and Humber), and an area may be split in two and sent into two different groups. The Met Office states the rule plainly: sea areas can be grouped if conditions will be similar, and the groupings vary according to the weather.
5 · How many words does it have?
The thirty-one area forecasts across 2,214 bulletins run to words. They use different ones. Sixty-eight of those carry ninety-nine per cent of the text.
Every word type in the corpus, sized and lit by how often it occurs. The dim ones at the end are the tail.
The tail is mostly the language failing, not growing
6 · Where the weather is, according to the words
A count of what the forecast said, not of what the weather did. The Met Office publishes its own verification of the shipping forecast against observations, going back more than a decade; this is not that, and is not a substitute for it. It is a portrait of the document.